DRGP-(V)SS(-IP-1/2)

(licenced under MIT licence)

We seperated the code into three parts for the model cases DRGP-(V)SS, DRGP-(V)SS-IP-1 and DRGP-(V)SS-IP-2. DRGP-(V)SS belongs to the folders: python, matlab; DRGP-(V)SS-IP-1 belongs to the folders: python_ip_true, matlab_ip_true folder. DRGP-(V)SS-IP-2 belongs to the folders: python_ip, matlab_ip folder.

In the following we describe the stages for initialization, optimization/training and testing/simulation for the DRGP-(V)SS cases. The DRGP-(V)SS-IP-1/2 cases are fully analog with respect to the different folder names python_ip, matlab_ip.

The initialization, optimization/training and testing/simulation code is split into three parts:

1. matlab initialization of the data for python
2. python optimization/training
3. matlab testing/simulation of the model

Steps to go:

1. • in \DRGP-VSS\matlab open the calibrate.m and change your prefered setting
   • choose create_data_python = 'on'; simulation = 'off'; to create the preprocessed data for python
   • Run STARTME.m
   • the created file 'load_python.mat' was saved in \DRGP-VSS\python\data
2. • in \DRGP-VSS\python open 'DEEPvSSGP_STARTME_SV1.py' (DRGP-VVS) or 'DEEPvSSGP_STARTME_SV2.py' (DRGP-SS)
   • choose a specific filename to load (dataset = 'load_python' or other)
   • if you have chosen non_rec = 'on' in matlab, specify non_rec = 1; in python, otherwise 0
   • choose the parameters to optimize and the amount of iterations
   • Run 'DEEPvSSGP_STARTME_SV1.py' (DRGP-VVS) or 'DEEPvSSGP_STARTME_SV2.py' (DRGP-SS)
   • You will be asked to enter something to define the name for the file to save
   • for the first time theano will compile the code before it will start to optimize
   • the optimized parameter-file is saved in \DRGP-VSS\matlab\data_optimized
3. • in \DRGP-VSS\matlab open the calibrate.m and choose the same setting as in 1. but choose create_data_python = 'off'; simulation = 'on' and specify filename_optimized_parameters = '...';
   • Run STARTME.m and wait for the results

We also offer some of our achieved results in \DRGP-VSS\matlab\data_optimized. The raw data is available in \DRGP-VSS\matlab\data_raw.

New:

We further added the Latent variable model (LVM) for data-dimensionality reduction with our GP approximations (V)SS, (V)SS-IP-1 and (V)SS-IP-2. The files can be found in python, python_ip and python_ip_true. We added two data-sets in python...\data: data_usps.mat and data_oil.mat. They consist of the initialization of our model parameters and the PCA initialization of the data.

---

---

Our code relies on some code snippets of VSSGP (https://github.com/yaringal/VSSGP). This means we took just these parts of the code which brings our new algorithm/method to work with theano/python. The main algorithm is new.

For these code snippets we include:

Copyright 2015 Yarin Gal

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.